System and method for selective treatment of crops using machine vision

Abstract

There is provided a system for customized application of herbicides, comprising: a processor(s) executing a code for: feeding test images corresponding to a target agricultural field into a machine learning model trained on a training dataset of sample images of sample agricultural field(s) labelled with ground truth of weed parameters, selecting specific weed parameter(s) of according to performance metric(s) of the model, setting up instructions for triggering application of a first herbicide to a portion of the target agricultural field in response to an outcome of the model indicating likelihood of the specific weed parameter(s) being depicted in an input image of the portion of the target agricultural field, and setting up instructions for triggering application of a second herbicide to the portion of the target agricultural field in response to the outcome of the model indicating non-likelihood of the specific weed parameter(s) being depicted in the input image.

What is claimed is: 
     
       1. A system for dynamic application of herbicides to a target agricultural field, comprising:
 at least one hardware processor executing a code configured for:
 running one or more machine learning models to detect weeds, wherein the one or more machine learning models is/are trained on at least one training dataset comprising a plurality of records, wherein each record includes at least one sample image depicting at least one weed and at least one ground truth label indicating a plurality of weed parameters; 
 obtaining an indication of a plurality of specific weed parameters of weeds present in the target agricultural field; 
 selecting a predicted weed parameter from the plurality of specific weed parameters, the predicted weed parameter selected when at least one performance metric of the one or more machine learning models for the predicted weed parameter is above a threshold; 
 generating instructions for loading a first herbicide into a spot spraying tank set for spot spraying of the weeds, the first herbicide selected for targeting first weeds present in the target agricultural field, the first weeds classified by the one or more machine learning models as having the selected predicted weed parameter; and 
 generating instructions for loading a second herbicide into a broadcast spraying tank set for broadcast spraying of the weeds, the second herbicide selected for targeting second weeds present in the target agricultural field, the second weeds classified by the one or more machine learning models as not having the selected predicted weed parameter. 
 
 
     
     
       2. The system of  claim 1 , wherein the code is further configured for receiving a user input that defines the threshold. 
     
     
       3. The system of  claim 1 , wherein:
 the first weeds comprise one or more first weed species, and 
 the second weeds comprise one or more second weed species that is/are different that the one or more first weed species. 
 
     
     
       4. The system of  claim 1 , wherein:
 the threshold comprises a size threshold, 
 the first weeds are classified as being larger than the size threshold, and 
 the second weeds are classified as being smaller than or equal to the size threshold. 
 
     
     
       5. The system of  claim 4 , wherein the first and second weeds comprise a same weed species. 
     
     
       6. The system of  claim 4 , wherein the size threshold corresponds to a growth stage. 
     
     
       7. The system of  claim 1 , wherein the threshold is set to differentiate a detection of the first weeds that are not visually similar to a ground and a detection of the second weeds that are visually similar to the ground. 
     
     
       8. The system of  claim 1 , wherein the threshold is set to differentiate a classification of the first weeds that are not visually similar to desired crops and a classification of the second weeds that are visually similar to the desired crops. 
     
     
       9. The system of  claim 1 , wherein the second weeds are of an uncertain species and/or an uncertain growth stage. 
     
     
       10. The system of  claim 1 , wherein the code is further configured for computing the at least one performance metric of the one or more machine learning models for each one of the plurality of weed parameters by analyzing a plurality of outcomes obtained by feeding the plurality of records into the one or more machine learning models. 
     
     
       11. The system of  claim 1 , wherein the at least one performance metric includes an accuracy of a classification and/or a detection by the one or more machine learning models for each of the plurality of weed parameters. 
     
     
       12. The system of  claim 1 , wherein each record includes a second ground truth label indicating at least one field parameter. 
     
     
       13. The system of  claim 12 , wherein the at least one field parameter includes a geographical location, a season, a phase during an agricultural growth cycle, a soil type, a tilled status of a soil, a weather, and/or a desired crop being grown. 
     
     
       14. A system for dynamic application of herbicides to a target agricultural field, comprising:
 at least one hardware processor; 
 a memory storing:
 one or more trained machine learning models configured to detect and/or classify weeds, the one or more machine learning models having been trained with at least one training dataset comprising a plurality of records, wherein each record includes at least one sample image depicting at least one sample weed and at least one ground truth label indicating at least one weed parameters; and 
 computer-readable instructions configured to be executed by the at least one hardware processor to cause the at least one hardware processor to:
 feed test images of an agricultural field into the one or more trained machine learning models; 
 determine an accuracy of a detection and/or a classification of test weeds in the test images; 
 produce a first output that indicates a first herbicide to load into a spot spraying tank for spot spraying of the weeds, the first herbicide selected for targeting first weeds present in a target agricultural field for which the accuracy is higher than a threshold; and 
 produce a second output that indicates a second herbicide to load into a broadcast spraying tank for broadcast spraying of the weeds, the second herbicide selected for targeting second weeds present in the target agricultural field for which the accuracy is lower than or equal to the threshold. 
 
 
 
     
     
       15. The system of  claim 14 , wherein the test images depict and/or represent the target agricultural field. 
     
     
       16. The system of  claim 14 , wherein:
 the computer-readable instructions further cause the at least one hardware processor to receive an input corresponding to an expected weed spectrum, and 
 the first and second outputs are produced based at least in part on the expected weed spectrum. 
 
     
     
       17. The system of  claim 16 , wherein the computer-readable instructions further cause the at least one hardware processor to automatically determine the expected weed spectrum using the one or more trained machine learning models and the test images. 
     
     
       18. The system of  claim 16 , wherein the input is a manual input received from a user interface. 
     
     
       19. The system of  claim 16 , wherein the computer-readable instructions further cause the at least one hardware processor to automatically determine the expected weed spectrum by analyzing a database storing weed data of the target agricultural field. 
     
     
       20. The system of  claim 19 , wherein the weed data comprise historical data and/or current data.